328 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

120 Gbit/s Over 500-km Using Single-BandPolarization-Multiplexed Self-Coherent

Optical OFDMBrendon J. C. Schmidt, Zuraidah Zan, Liang B. Du, and Arthur J. Lowery, Fellow, IEEE

Abstract—In this paper, we experimentally demonstrate asingle-band self-coherent polarization-multiplexed optical orthog-onal frequency-division multiplex system with a raw data rateof 120 Gbit/s. The transmitter uses a novel RF structure thateliminates the need for RF mixers and optical filters. The receiveruses a novel architecture where the optical carrier is filtered andamplified for self-coherent detection. The receiver is polarizationdiverse and allows for the usual frequency guard band between thecarrier and the sideband to be reduced in width, thus increasingspectral efficiency. Using two commercial 20 GS/s arbitrary-wave-form generators to generate a single information-carrying bandper polarization, we achieve a raw data rate of 120 Gbit/s over 500km of standard single-mode fiber.

Index Terms—Chromatic dispersion, compensation, directdetection, orthogonal frequency-division multiplexing (OFDM),optical single sideband (OSSB), polarization multiplexing, self-co-herent.

I. INTRODUCTION

T HE rapidly increasing traffic demands in long-haul op-tical fiber communications have caused considerable in-

terest in alternative modulation formats that can efficiently usethe capacity of standard single-mode fiber (SSMF) [1]. As 100Gbit Ethernet (100 GbE) is set to become the next standard forIP transport, next-generation optical systems need to be ableto transport 100 GbE in a single wavelength [2]. Differentialquadrature phase-shift keying (DQPSK) using polarization-di-vision multiplexing (PDM) and either direct or coherent detec-tion [3] are possible solutions. Recent demonstrations of PDM-8PSK [4] using coherent reception and offline DSP achieveda high tolerance for chromatic dispersion (CD) and polariza-tion-mode dispersion (PMD).

Another modulation format that has attracted considerableinterest for long-haul transmission is optical orthogonal fre-quency-division multiplexing (O-OFDM) [5]–[7]. This formathas many advantages: 1) it can adaptively compensate for very

Manuscript received June 01, 2009; revised August 05, 2009 and August 23,2009. First published October 09, 2009; current version published February 01,2010. This work was supported by the Australian Research Council’s DiscoveryFunding Scheme (DP 0772937).

The authors are with the Department of Electrical and Computer Systems En-gineering, Monash University, Clayton, Vic. 3800, Australia (e-mail: brendon.schmidt@eng.monash.edu.au).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/JLT.2009.2033613

high levels of CD [5] and PMD [8]; 2) the computational com-plexity scales well with increasing data rate [1]; and 3) the signalhas a well-defined optical spectrum with high spectral efficiency[9], [10]. Recently, there have been several demonstrations thatshow the potential of O-OFDM for next-generation 100-Gbit/ssystems. O-OFDM using coherent reception (CO-OFDM) andPDM has been used to achieve 100 GbE data rates over 1000 km[11], [12]. In both demonstrations, the wideband optical signalwas generated by multiplexing a number of narrower signalbands (subbands) together. In [11], the bands were generated inthe optical domain by modulating a comb of laser lines, similarto ‘‘coherent wavelength division multiplexing (WDM)’’[13]but with OFDM modulation of each line. In [12], the bandswere generated in the electrical domain and then upconverted toselected intermediate RFs before optical modulation. The use ofmultiple bands allows the sampling rate of the digital-to-analogconverters (DACs) to be reduced, and reduces some OFDMoverheads [12]. However, these benefits come at the cost oftransmitter complexity. Both systems require multiple DACs;optical multiplexing requires multiple optical modulators andRF multiplexing requires multiple electrical mixers. There arealso ‘‘all-optical’’ versions of O-OFDM [14], using no digitalprocessing at the transmitter. These similarly require multipleoptical modulators.

In this paper, we demonstrate a 120-Gbit/s self-coherentO-OFDM (SCO-OFDM) system that uses only one signalband per polarization; a 100-Gbit/s version of this system waspresented at the Optical Fiber Communication Conference andExposition (OFC) 2009 [15]. The system does not use subbandsin either the RF or optical domains. The use of direct detection[16] means that a laser “local oscillator” is not required at thereceiver. Single-band schemes have far simpler RF and opticalpaths in the transmitter but are limited by the performance ofthe transmitter DACs. We have aimed at achieving the max-imum performance from second-generation arbitrary waveformgenerators (AWGs) as of May 2009.

Our system uses a polarization-diverse receiver [17], unlikealternative direct-detection receivers presented at the OFC/Na-tional Fiber Optic Engineers Conference (NFOEC) 2009 post-deadline sessions [18]. This means that the performance of PDMis not degraded in high-PMD optical links. The system has alsoa number of novel features developed by our group, includinga transmitter with no RF mixers or optical filters [19], carrierboost to improve receiver sensitivity [20], and the polarization-diverse receiver itself [17]. The system uses a reduced ‘‘gap’’[21] between the transmitted carrier and the OFDM sideband,

0733-8724/$26.00 © 2010 IEEE

SCHMIDT et al.: SELF-COHERENT OPTICAL OFDM 329

Fig. 1. Desired output spectrum of the transmitter.

which is possible through the use of balanced detectors at thereceiver. This reduced gap improves the spectral efficiency dra-matically, compared with a direct-detection O-OFDM system(DDO-OFDM) [16], but retains its tolerance to phase noise fromthe transmitter’s laser. In contrast, CO-OFDM is sensitive tolaser phase noise unless phase compensation techniques [22]or narrow-linewidth lasers and pilot tones [23] in the OFDMsymbol are used.

The paper is organized as follows. Section II describes thekey features of the system. Section III provides the experimentaldetails. In Section IV, we present and discuss the results andcompare them to the results from simulation. In Section V, wepresent our conclusions.

II. SYSTEM DESIGN THEORY

This section covers the design concepts used in the system toachieve a wide modulation bandwidth with low sample rates atthe transmitter and polarization diversity at the receiver.

A. Transmitter Design

SCO-OFDM, like DDO-OFDM, requires an optical carrier tobe transmitted along with the OFDM subcarriers. This carriermixes with the optical subcarriers at the photodiode to producea set of electrical subcarriers. Since the carrier and the subcar-riers are derived from the same laser source, the phase noise ofthe source (due to laser linewidth) will cancel upon detection,provided there is little differential delay between the carrier andthe subcarriers.

Fig. 1 shows the desired optical spectrum for a polarization-multiplexed DDO-OFDM transmitter. This comprises subcar-rier bands in orthogonal polarizations, together with an opticalcarrier in one (or other) of the polarizations. In this figure, theOFDM subcarriers are modulated with 16-quadratic-amplitudemodulation (QAM) in the center of the subcarrier band and4-QAM at the edges of the subcarrier band, though other ar-rangements are possible. The source could be a tunable laser (toallow the transmitter to be adjusted to any wavelength). A vir-tual carrier is created by frequency-shifting the tunable laser lineto one side of the subcarrier band [19]. A ‘‘gap’’ is left in thespectrum. In most DDO-OFDM systems [21], the gap ensuresthat intermixing products between subcarriers, caused by pho-todetection, fall below the subcarriers’ frequency band: in thereceiver design presented later, the gap’s purpose is to allow thecarrier and subcarrier bands to be separated using optical filters.

The spectrum of Fig. 1 can be generated by applying theinphase (I) and quadrature (Q) components of the electricalOFDM signal to a complex optical modulator (also known as

Fig. 2. Optical OFDM transmitter for a single polarization.

a QPSK modulator, optical single-sideband modulator, or cas-caded triple Mach–Zehnder interferometric (MZI) modulator).This arrangement is shown in Fig. 2. The electrical OFDMsignal is generated by presenting parallel data to a set of QAMmodulators. The QAM modulators’ outputs then become the(frequency domain) inputs to an inverse fast Fourier transform(IFFT). The output of the IFFT is a superposition of subcarriers,each corresponding to a frequency input of the IFFT. These areconverted to a complex serial data stream and a cyclic prefix(CP) is added.

The optical modulator is biased at null power so that the op-tical field is proportional to the electrical drive voltage [24].This ensures a one-to-one mapping of the optical and electricalspectra, both at the transmitter and the receiver. This is neces-sary to enable fiber dispersion to be equalized electrically on asubcarrier-by-subcarrier basis. The virtual carrier is generatedby frequency-shifting the original line. This can be achieved byadding sine and cosine waveforms to the two inputs of the com-plex optical modulator, so that it operates as a single-sidebandmodulator to create a single sideband comprising the virtual car-rier. The frequency shift is equal to the frequency of the oscil-lator, . The power in the optical carrier can be controlledsimply by the amplitude of the RF waveform and the frequencygap by adjusting its frequency. This is very convenient experi-mentally and allows a flexible transmitter module to be manu-factured.

An advantage of the transmitter design in Fig. 2, over previousdesigns, is that it provides the maximum possible optical sub-carrier-band bandwidth for a given sample rate [19]. Neglectingthe narrow guard bands required for antialiasing filtering, thebandwidth of the optical subcarrier band can equal the samplerate of the DACs. For example, 20 GS/s DACs will give a totalsubcarrier bandwidth approaching 20 GHz, which could sup-port close to 40-Gbit/s (per polarization) using 4-QAM modu-lation, or 80 Gbit/s per polarization with 16 QAM. These datarates do not include forward error correction (FEC) overheadsor the CP overhead. Alternative approaches generate the virtualcarrier by setting the inputs of the IFFT appropriately [25], soone half of the double-sided spectrum is sacrificed for the gapand the carrier, or by frequency-shifting the electrical subcarrierband away from the laser line using RF mixers [21]. Schemesusing RF mixers are undesirable because they are expensive andcan introduce nonlinear distortion to the signal, and have lim-ited bandwidths. Also, it is desirable to place the subcarrier bandaround dc of the modulator’s electrical response (at positive and

330 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

Fig. 3. Dual-polarization optical OFDM receiver.

negative frequencies around dc), because this is where it hasits maximum efficiency and flatness. RF mixers would use themodulator at some intermediate frequency, and the IFFT gener-ated spectrum uses only positive frequencies for the subcarriers.The virtual carrier is placed at a very high frequencyin our scheme. The low efficiency of the optical modulator atthis frequency is not important, because powerful narrowbandhigh-frequency RF amplifiers are readily available to compen-sate for this low efficiency.

Two signal polarizations can be generated by combining theoutputs of two transmitters, each with a polarization beam com-biner, as shown in Fig. 2. A desirable simplification is to removethe carrier generation feature from one transmitter; otherwise,the two carriers will have a random phase shift, and thus create acarrier with a random polarization state. One carrier is easily re-moved by removing one transmitter’s sinusoidal generator. An-other desirable simplification is to use a single tunable laser forboth polarizations, as this will ensure that the bands of both po-larizations have phase fluctuations correlated with the commoncarrier, so these phase fluctuations will cancel upon photodetec-tion.

B. Receiver Design

We use an improved version of the high-performance receiverof [17], using balanced photodetectors rather than single-endedphotodetectors. This receiver reduces the sensitivity limitationsof conventional direct-detection receivers [26], allows polariza-tion multiplexing and PMD compensation, and maintains direct-detection receiver’s insensitivity to laser linewidth by using self-carrier extraction [27], [28]. The balanced photodetection re-jects unwanted spectral components, including the intermixingof subcarriers with other subcarriers. This allows a reduced fre-quency gap, and thus gives an increased spectral efficiency.

The topology of the receiver is shown in Fig. 3. The first stage(filter/demux) comprises two optical filters to separate the car-rier and sideband spectral components. The filter also providesamplified spontaneous emission (ASE) bandwidth limiting toreduce shot noise. The separated carrier and sideband signalsare then optically pre-amplified and drive the inputs of the po-larization-diverse optical hybrid. The hybrid has an internal po-larization beam splitter on the signal input and an internal linearpolarizer on the LO input. The linear polarizer ensures equal car-rier power for each polarization and the polarization controllermaximizes carrier output.

The hybrid provides balanced inphase outputs for the andpolarizations: the receiver is a heterodyne rather than a ho-

modyne design, so the I and Q components are recovered usingRF mixers. The and polarization outputs of the optical hy-brid are detected with balanced photodetectors, in contrast to[17]. These strongly suppress subcarrier subcarrier mixing

Fig. 4. Transmitter as implemented using commercial components.

products [26], which would normally fall in the frequency gapin a direct-detection design. Thus, the width of the gap can bereduced to below the bandwidth of the subcarrier band withoutpenalty, provided the optical filters are sharp enough. The bal-anced photodetectors also reject ASE ASE components. Theoptical preamplification serves two functions: it increases thesignal powers to minimize shot noise upon detection; boostingthe carrier power [20] would allow unbalanced detection to beused [29], but in this design, it helps to reduce the impact ofresidual subcarrier subcarrier mixing products caused byimbalances in the balanced detection. The detected outputs arethen amplified and downconverted to the electrical basebandwith two wideband electrical mixers. The I and Q outputs fromeach mixer are then low-pass filtered and captured by the fourinputs of a digitizer. The digitized waveforms are convertedto complex parallel data blocks and the CP is removed. TheIQ compensation, channel estimation, and polarization demul-tiplexing are all performed in the DSP block, before the QAMsymbols are demodulated into binary data.

III. EXPERIMENTAL DETAILS

The transmitter and the receiver described in Figs. 2 and 3were constructed using commercial off-the-shelf components.This required some changes to their designs. Also, to savebuilding two transmitters (one for each polarization), theand polarization signals were generated using one modu-lator, then combined with a relative delay to decorrelate them.This section describes the implementation of the transmitter,receiver, and fiber link.

A. Implemented Transmitter

Fig. 4 shows the transmitter as implemented experimentally.The QAM coders, IFFT, and CP addition are all generated of-fline using MATLAB. The inphase component of the gener-ated waveform is loaded into one Tektronix AWG7102 and thequadrature waveform into the other. The AWGs are used ininterleaved and ‘‘return-to-zero’’ modes to provide an outputsample rate of 20 GS/s each, with an analog signal bandwidth ofapproximately 6 GHz ( dB) and 10 GHz ( dB). Low-passfilters (LPFs) with sharp 10-GHz cutoffs are used to reduce theimage band caused by the sampling process.

The outputs of the LPFs are amplified with 20-GHzbandwidth microwave amplifiers and fed to a Sumitomo

SCHMIDT et al.: SELF-COHERENT OPTICAL OFDM 331

T-SBXI.5-20P 40-Gbit/s DQPSK optical modulator (a “com-plex optical modulator,” COM). The source laser is a Pho-tonetics Tunics external-cavity laser tuned to 1550 nm. Anoptical amplifier is used before the modulator to compensatefor the low output power of this particular laser. The output ofthe modulator is split into two paths: one with a delay of either54 or 64 ns, depending on the length of the CP. Polarizationcontrollers (PCs) adjust the polarizations of the paths at theinput to a polarization beam combiner (PBC). This combinationemulates two separate transmitter cards, each supplying signalswith orthogonal polarizations to one another.

The carrier could be generated in a similar manner to thatshown in Fig. 2; however, the arrangement to generate twosignal polarizations would cause the carrier and a delayed androtated version of itself to be transmitted. We found that thiscauses the carrier at the receiver to change polarization rapidly(because the ‘‘blue’’ delay path and the ‘‘red’’ straight pathwill fluctuate in relative phase length due to vibrations), andthe polarization at the receiver cannot be manually controlled.Thus, the carrier is actually generated by a separate modulatoracting as a frequency shifter and added to the output of thePBC. This removed all the problems of a rapidly changingcarrier polarization.

It is important to generate the I and Q waveforms with a pre-cise phase relationship; otherwise, there will be leakage fromthe upper sideband to the lower sideband of the subcarriers,causing a large penalty. To achieve this, the AWGs are synchro-nized using a common clock and trigger source. This achievesa long-term mean offset of 1 ps and an rms timing jitter of 4ps, as measured on a Tektronix 72004 DSO. The AWGs’ rolloffs, and large frequency response ripples in ‘‘interleaved’’ and‘‘zeroed’’ modes are compensated by multiplying the inverseof each AWG’s channel response with the QAM transmissiondata in the frequency domain, before the IFFT. This leads toa flat response, but the preemphasis causes a loss of DAC res-olution. This loss, combined with the SNR penalty caused byusing the DAC in ‘‘zeroed mode’’ (which is ‘‘return to zero’’after each sample) causes the electrical SNR at the transmitter tobe a major impairment to the bit error rate (BER) performanceof the system. The electrical SNR impairments can be reducedby using the well-known bit loading techniques [30]. For thisreason, a combination of 16 QAM was used for the center sub-carriers and 4 QAM for the higher frequency subcarriers.

B. Fiber Link

The fiber plant uses seven spans of Corning SMF-28e:1 50 km, 5 80 km, and 1 50 km for a transmissiondistance of 500 km with no dispersion compensation. Light-Waves2020 erbium-doped fibre amplifiers (EDFAs) are usedbetween each span. The input power into each span was set toapproximately 0 dBm.

C. Implemented Receiver

The output of the final span is optically amplified and fedto the common port of a programmable optical filter (Finisar‘‘Waveshaper 4000E’’). The carrier and sideband spectral com-ponents are separated by the optical filters and routed to separateoutput ports. A filter with a 12-GHz full-width at half-maximum

TABLE IKEY PARAMETERS OF THE EXPERIMENTAL SYSTEM

(FWHM) is used to separate the carrier; a 20-GHz filter extractsthe sideband. The separated carrier is amplified to 15 dBm andthe sideband to only 3 dBm to provide the equivalent of carrierboosting [20]. The amplified carrier and sideband are fed to theinputs of a Kylia MINT 2 8 optical hybrid. This can produceI and Q outputs for both the and polarizations. However,as the carrier frequency is offset from the sideband, the hybridwas used as a heterodyne receiver, so only the inphase outputsfor the and polarizations are required.

Polarization control is required between the output of thecarrier filter and the LO input of the optical hybrid to maximizethe output carrier level. A manual controller was used in theseexperiments; however, a practical implementation would re-quire a closed-loop controller. The outputs of the hybrid are fedinto two U T 40-GHz balanced photodetectors, one for eachpolarization. The photodetectors produced an electrical outputwith an intermediate frequency of 20 GHz, over a band from ap-proximately 10 to 30 GHz. The outputs are each amplified with40-GHz RF amplifiers (Marki A-0040) and then downconvertedto the complex baseband with wideband IQ mixers (Marki). In[15], two independent 20-GHz sinewave generators were used:one at the transmitter for virtual carrier generation and the otherat the receiver for electrical downconversion of the photodiodeoutputs. However, a common 20-GHz generator with splitoutputs for the transmitter and receiver ends did not affect theresults, and all of the experiments in this paper use a single20-GHz generator. The downconverted andsignals are sampled in real time by the four channels of aTektronix 72004 20-GHz 50-GS/s oscilloscope. The samplesare then analyzed in MATLAB, which includes algorithmsfor IQ imbalance correction [12], polarization demultiplexing,channel equalization, and QAM decoding.

IV. EXPERIMENTAL RESULTS

A combination of 4-QAM and 16-QAM subcarrier modula-tion is used to demonstrate the performance of the system. Araw data rate of 120 Gbit/s can be transmitted over a distanceof 500 km of SSMF. This data rate is enough to cover the over-heads for 100 GbE transmission. Table I shows the subcarrierallocations, bandwidths, and data rates contributed by the twomodulation levels.

The experimental system operates using frames of data,which the AWGs repeat over and over from their memories.The frames are designed so that the signals can be equalizedusing the training within a single frame, which is a limitationimposed by offline processing. A frame comprises a trainingsequence of 60 OFDM symbols and a data payload of 150symbols (467 000 random bits). The long training sequence

332 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

reduces the average data rate to 85 Gbit/s. This overhead canbe reduced by using longer data payloads and using runningaverages over multiple frames to reduce the required length ofeach training symbol [12]. The techniques proposed in [31]would reduce the training overheads further.

The training is required to find the matrix coefficients forboth IQ compensation and polarization demultiplexing at the re-ceiver. As described in [12], the training for a single transmittersystem with a delay line for decorrelation requires a few modi-fications. To enable polarization demultiplexing at the receiver,the training sequence begins with 30 OFDM training symbolsinterleaved with 30 empty symbols, then develops as follows:1) the training components that travel directly to the inputof the polarization beam combiner become the training sym-bols for the polarization and 2) the training components thattravel via the delay line to the input of the polarization beamcombiner become the training symbols for the polarization.The delay line causes the training symbols to occupy the pre-viously empty symbol slots, and the training sequence after thepolarization combiner becomes a set of 60 interleaved andtraining symbols. The known training data and the receivedand data can then be used at the receiver to demultiplex the

and data.The data content of the 30 training symbols is selected to

give low peak-to-average power ratios (PAPRs). The symbolsare scaled to just below the clipping level for the payload data.This scaling increases the training symbol energy by 2.3 dB andhelps offset the 3-dB power loss in the training sequence causedby the empty symbol slots. The training symbols have no clip-ping noise and their increased energy helps reduce impairmentscaused by DAC noise.

A. Optical Spectrum

Fig. 5 shows the optical spectrum obtained with an Agilenthigh-resolution spectrometer (HRS) after 500-km of fiber.Careful adjustments of the modulator biasing, microwave pathlengths, and attenuator pads were required to suppress theoriginal laser line (in the middle of the subcarrier band) and theimage of the virtual carrier that lies above the subcarrier band.The HRS’s spectrum shows the 4-QAM (outer) and 16-QAM(inner) subcarrier bands, similar to Fig. 1. The spectrum alsoshows a small residual of the original laser line and virtualcarrier image (30-dB suppression) due to small amplitudemismatches in the I and Q drive levels. Beneficially, the imageof the virtual carrier falls away easily from the sideband, unlikeother virtual carrier schemes.

B. Electrical SNR

Fig. 6 shows the received electrical SNR/bit for the 120 Gbit/stransmission experiment through 500-km SMF, with a 25% CP.The measured OSNR was 25.8 dB (0.1 nm) and the BER was

. The subcarrier’s ordering is the same as the orderingin the optical spectrum in Fig. 5. In most cases, the amplitude forthe optical spectrum is a useful guide for the received SNR/bit,and areas of significant difference suggested that there may be aproblem with the receiver. For example, the drooping responsefor increasing negative frequencies is caused by an overlap in thepassbands of the receiver optical filters that separate the carrier

Fig. 5. Optical spectrum after 500-km of fiber, obtained with a high-resolutionspectrophotometer (20-MHz resolution).

Fig. 6. Electrical SNR/bit �� �� � measured for the 120 Gbit/s transmissionexperiment with 500 km of SSMF, for an OSNR of 25.8 dB (0.1 nm) and ������ BER. An SNR/bit of 6.8 dB is required for a BER of ���� for 4 QAMand 10.6 dB is required for 16 QAM.

and the sideband. This dip can be reduced by offsetting the fre-quency of the carrier filter, but this leads to less carrier power andincreased noise at the input to the optical hybrid. In the exper-iments, a compromise between filter overlap and carrier outputis used to minimize the error rate.

C. Received and Equalized Constellations

Fig. 7 shows the constellations for the 4-QAM and 16-QAMsubcarriers after 500-km, with the and polarizations plottedin red and blue, respectively. The 4-QAM and 16-QAM constel-lations have similar spreads compared with their separations, in-dicating that they will offer similar bit error ratios.

D. BER Versus OSNR

Fig. 8 shows the BER versus OSNR plots for the opticalback-to-back and 500-km transmission measurements, and sim-ulation results from VPItransmissionMaker. The OSNR wasmeasured with a calibrated Agilent 86142B optical spectrumanalyzer set to a resolution bandwidth of 0.5 nm (to capture thewhole signal power) and then rescaled to 0.1 nm (the standardbandwidth for OSNR calculations). The slightly lower power(0.7 dB) in the training sequence is accounted for, so that the

SCHMIDT et al.: SELF-COHERENT OPTICAL OFDM 333

Fig. 7. Constellations for 4-QAM and 16-QAM subcarriers after 500 km offiber. Note red points (�-polarization) are sometimes plotted over blue points(� -polarization).

Fig. 8. BER versus OSNR plots. The line indicates simulated results, the tri-angles are optical back-to-back measurements, the stars indicate 500 km trans-mission measurements and are an average of the results of the � and � polar-izations.

OSNR accurately represents the required OSNR for a systemwith infrequent training.

The plots for the measured data show a tendency toward aBER floor. This is due to limitations in the AWGs’ DAC perfor-mance: to increase the sample rate and the analog bandwidth, theAWGs are run in ‘‘interleaved’’ and ‘‘zeroed’’ mode. This modeprovides increased bandwidth but at the expense of increasedspurious noise and a halving of the output amplitude. The DACperformance is also reduced by the strong preemphasis requiredat high frequencies.

There is a considerable difference in the BER performance ofthe and polarizations, shown in the optical back-to-backresults in Fig. 8. This is because the data in the polarizationhave a delay of one symbol period relative to the carrier. Thisdecorrelates the laser phase noise between the carrier and thesubcarriers in the polarization, so the laser phase noise doesnot cancel effectively upon photodetection. In Fig. 8, the penal-ties due to the laser phase noise increase from 2 to dB asthe polarization’s BER floor is approached.

The back-to-back performance of the polarization waswithin 2.5 dB of the simulated result for a BER of . Therequired OSNR in the simulation, for the generally 16-QAMsignal, was 21.9 dB and the experimental result was 24.3 dB.The simulation is based on the same topology as the experi-mental system and uses perfect components with one exception:the simulations model nonideal optical filters with bandwidthssimilar to the optical filters used in the experiment. The opticalfilter bandwidths, offset, and order have a significant effect onthe OSNR requirements, and the simulation filter orders wereincreased to third order to ensure conservative comparisons.The experimental OSNR for 500-km transmission using a 25%CP required 25.8 dB, compared to a simulated 22.2 dB at aBER of . A large part of the increased penalty from2.5 to 3.6 dB is due to the transmission BER being an averageover the BERs of the and polarizations, when it mightbe expected that the channel could be improved to thechannel’s performance if the additional delay is removed. Theexperimental OSNR for 500-km transmission using a 5% CPwas 26 dB for a BER of . The decreased performancewas due to experimental error that led to a CP of less than 5%.

V. CONCLUSION

We have demonstrated a 120-Gbit/s self-coherent opticalOFDM system capable of transmitting more than 500-kmusing readily available commercial components and test equip-ment. Only one electrical and one optical band are used atthe transmitter, rather than multiplexing together individualelectrical bands, optically or electrically. Several advanceddesign features are combined to obtain this performance. Theoptical transmitter uses a virtual carrier scheme to maximize thequality of the transmitted signal by placing the subcarrier bandsat the center of the modulator’s response. The virtual carriercan be adjusted finely to tune its frequency and amplitude byadjusting one RF signal.

At the receiver, a polarization-diverse self-coherent systemwas developed. This uses an optical filter to separate the car-rier from the signal band, then a polarization diverse coherentheterodyne receiver with PMD compensation and polarizationdemultiplexing. Balanced photodiodes were used. These cancelsubcarrier subcarrier noise, which is usually a limitation to

334 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 4, FEBRUARY 15, 2010

the spectral efficiency of direct-detection systems, as a gap hasto be left in the spectrum for them to fall in, upon photodetec-tion. The new limitation is the performance of the optical filter.In the demonstration, we could reduce the gap from the signalbandwidth (19 GHz) to 8 GHz without penalty.

The OSNR requirement was close to a near-perfect systemsimulation. Compared with a coherent system, some additionalOSNRs are required because the carrier is also transmittedalong with the subcarriers. This adds 3 dB to the OSNR re-quirement, although carrier suppression [20] could reduce thispenalty. The use of 16-QAM introduces a theoretical 3.7-dBOSNR penalty over 4-QAM. Thus, the performance is expectedto be within 7-dB of coherent systems. For comparison, a recentexperimental 4-QAM five-band 107 Gbit/s coherent O-OFDMsystem required 17 dB OSNR [11]. Improved DACs wouldallow our method to use 4-QAM, potentially reducing ourOSNR requirement to within 4 dB of the coherent sensitivity.

REFERENCES

[1] H. Bülow, “Tutorial: Electronic dispersion compensation,” presented atthe Opt. Fiber Telecommun., Anaheim, CA, 2007, Paper OMG5.

[2] M. Cvijetic, “Towards 100 GbE introduction: Challenges and practicalaspects,” presented at the 10th Anniversary Int. Conf. Transparent Opt.Netw. 2008 (ICTON 2008), Athens, Greece, , Paper 1–4.

[3] C. R. S. Fludger, T. Duthel, D. van den Borne, C. Schulien, E. D.Schmidt, T. Wuth, J. Geyer, E. De Man, G.-D. Khoe, and H. de Waardt,“Coherent equalization and POLMUX-RZ-DQPSK for robust 100-GEtransmission,” J. Lightw. Technol., vol. 26, no. 1, pp. 64–72, Jan. 2008.

[4] J. Yu, D. Qian, T. Wang, G. Zhang, and P. D. Magill, “8� 114 Gb/s,25-GHz-spaced, PolMux-RZ-8 PSK transmission over 640 km ofSSMF employing digital coherent detection and EDFA-only amplifi-cation,” presented at the Opt. Fiber Commun., San Diego, CA, 2008,Paper PDP1.

[5] A. J. Lowery, L. Du, and J. Armstrong, “Orthogonal frequency divisionmultiplexing for adaptive dispersion compensation in long haul WDMsystems,” presented at the Opt. Fiber Commun. (OFC 2006), Anaheim,CA, , Paper PDP39.

[6] W. Shieh and C. Athaudage, “Coherent optical orthogonal frequencydivision multiplexing,” Electron. Lett., vol. 42, no. 11, pp. 587–588,May 2006.

[7] I. B. Djordjevic and B. Vasic, “Orthogonal frequency division multi-plexing for high-speed optical transmission,” Opt. Exp., vol. 14, pp.3767–3775, 2006.

[8] W. Shieh, “PMD-supported coherent optical OFDM systems,” IEEEPhoton. Technol. Lett., vol. 19, no. 3, pp. 134–136, Feb. 2007.

[9] H. Takahashi, A. Al Amin, S. L. Jansen, I. Morita, and H. Tanaka,“8� 66.8-Gbit/s coherent PDM-OFDM transmission over 640 km ofSSMF at 5.6-bit/s/Hz spectral efficiency,” in Proc. 34th Eur. Conf. Opt.Commun. 2008 (ECOC), pp. 1–2.

[10] H. Takahashi, A. A. Amin, S. L. Jansen, I. Morita, and H. Tanaka,“DWDM transmission with 7.0-bit/s/Hz spectral efficiency using8� 65.1-Gbit/s coherent PDM-OFDM signals,” presented at the Opt.Fiber Commun. (OFC 2009), San Diego, CA, , Paper PDPB7.

[11] W. Shieh, Q. Yang, and Y. Ma, “107 Gb/s coherent optical OFDMtransmission over 1000-km SSMF fiber using orthogonal band multi-plexing,” Opt. Exp., vol. 16, pp. 6378–6386, 2008.

[12] S. L. Jansen, I. Morita, T. C. W. Schenk, and H. Tanaka, “121.9-Gb/sPDM-OFDM transmission with 2-b/s/Hz spectral efficiency over 1000km of SSMF,” J. Lightw. Technol., vol. 27, no. 3, pp. 177–188, Feb.2009.

[13] A. D. Ellis and F. C. G. Gunning, “Spectral density enhancementusing coherent WDM,” IEEE Photon. Technol. Lett., vol. 17, no. 2,pp. 504–506, Feb. 2005.

[14] A. Sano, E. Yamada, H. Masuda, E. Yamazaki, T. Kobayashi, E.Yoshida, Y. Miyamoto, S. Matsuoka, R. Kudo, K. Ishihara, Y. Taka-tori, M. Mizoguchi, K. Okada, K. Hagimoto, H. Yamazaki, S. Kamei,and H. Ishii, “13.4-Tb/s (134� 111-Gb/s/ch) no-guard-interval co-herent OFDM transmission over 3 600 km of SMF with 19-ps averagePMD,” in Proc. 34th Eur. Conf. Opt. Commun. (ECOC), 2008, pp.1–2, Paper 1–2.

[15] B. J. C. Schmidt, Z. Zan, L. B. Du, and A. J. Lowery, “100 Gbit/s trans-mission using single-band direct-detection optical OFDM,” presentedat the Opt. Fiber Commun. (OFC 2009), San Diego, CA, , paper PD3.

[16] B. J. C. Schmidt, A. J. Lowery, and J. Armstrong, “Experimentaldemonstrations of electronic dispersion compensation for long-haultransmission using direct-detection optical OFDM,” J. Lightw.Technol., vol. 26, no. 1, pp. 196–203, Jan. 2008.

[17] B. J. C. Schmidt, A. J. Lowery, and J. Armstrong, “Impact of PMDin single-receiver and polarization-diverse direct-detection opticalOFDM,” J. Lightw. Technol., vol. 27, no. 14, pp. 2792–2799, Jul. 2009.

[18] D. Qian, N. Cvijetic, J. Hu, and T. Wang, “108 Gb/s OFDMA-PONwith polarization multiplexing and direct-detection,” presented at theOpt. Fiber Commun. (OFC 2009), San Diego, CA, , Paper PD5.

[19] B. J. C. Schmidt, A. J. Lowery, and L. B. Du, “Low sample rate trans-mitter for direct-detection optical OFDM,” presented at the Opt. FiberTelecommun. (OFC 2009), San Diego, CA, , Paper OWM4.

[20] A. J. Lowery, “Improving sensitivity and spectral efficiency in di-rect-detection optical OFDM systems,” presented at the Opt. FiberTelecommun. (OFC 2008), San Diego, CA, , Paper OMM4.

[21] A. J. Lowery and J. Armstrong, “Orthogonal frequency division mul-tiplexing for dispersion compensation of long-haul optical systems,”Opt. Exp., vol. 14, pp. 2079–2084, 2006.

[22] S. L. Jansen, I. Morita, N. Tadeka, and H. Tanaka, “20-Gb/s OFDMtransmission over 4 160-km SSMF enabled by RF-pilot tone phasenoise compensation,” presented at the Opt. Fiber Telecommun., Ana-heim, CA, 2007, Paper PDP15.

[23] X. Yi, W. Shieh, and Y. Ma, “Phase noise effects on high spectral effi-ciency coherent optical OFDM transmission,” J. Lightw. Technol., vol.26, no. 10, pp. 1309–1316, May 2008.

[24] B. J. C. Schmidt, A. J. Lowery, and J. Armstrong, “Experimentaldemonstrations of 20 Gbit/s direct-detection optical OFDM and12 Gbit/s with a colorless transmitter,” presented at the Opt. FiberCommun., Anaheim, CA, 2007, Paper PDP18.

[25] W.-R. Peng, X. Wu, V. R. Arbab, B. Shamee, L. C. Christen, J.-Y.Yang, K.-M. Feng, A. E. Willner, and C. Sien, “Experimental demon-stration of a coherently modulated and directly detected opticalOFDM system using an RF-tone insertion,” presented at the Opt. FiberCommun., San Diego, CA, 2008, Paper OMU2.

[26] A. J. Lowery, “Amplified-spontaneous noise limit of optical OFDMlightwave systems,” Opt. Exp., vol. 16, pp. 860–865, 2008.

[27] L. Xu, J. Hu, D. Qian, and T. Wang, “Coherent optical OFDM sys-tems using self optical carrier extraction,” presented at the Opt. FiberCommun. (OFC 2008), San Diego, CA, , .

[28] C. Xie, “PMD insensitive direct-detection optical OFDM systemsusing self-polarization diversity,” presented at the Opt. Fiber Commun.(OFC 2008), San Diego, CA, , .

[29] A. Carena, V. Curri, P. Poggiolini, and F. Forghieri, “Dynamic rangeof single-ended detection receivers for 100 GE coherent PM-QPSK,”IEEE Photon. Technol. Lett., vol. 20, no. 15, pp. 1281–1283, Aug.2008.

[30] S. T. Chung and A. J. Goldsmith, “Degrees of freedom in adaptivemodulation: A unified view,” IEEE Trans. Commun., vol. 49, no. 9,pp. 1561–1571, Sep. 2001.

[31] X. Liu and F. Buchali, “Intra-symbol frequency-domain averagingbased channel estimation for coherent optical OFDM,” Opt. Exp., vol.16, pp. 21944–21957, 2008.

Brendon J. C. Schmidt received the B.Eng. degree(first class hons) in electrical and computer systemsengineering from Monash University, Melbourne,Australia in 2006, where he is currently workingtowards the Ph.D. degree.

His research interests include OFDM in opticalcommunications applications.

SCHMIDT et al.: SELF-COHERENT OPTICAL OFDM 335

Liang Bangyuan Du was born in Shengyang, China,in 1985. He received the B.Eng. degree (Hons) andcommerce majoring in electrical and computer engi-neering and finance, respectively, from Monash Uni-versity, Melbourne, Australia, where he is currentlyworking towards the Ph.D. degree in the Departmentof Electrical and Computer Systems Engineering.

His research interests include optical orthogonalfrequency division multiplexing and fiber nonlin-earity.

Zuraidah Zan received the B.Eng. degree (Hons) incomputer and communication systems from Univer-siti Putra Malaysia, Malaysia, in 2003, and the M.Sc.degree in communication engineering from Univer-siti Kebangsaan Malaysia, Malaysia, in 2006y. She iscurrently working toward the Ph.D. degree at MonashUniversity, Melbourne, Australia with her major areaof research being the optical orthogonal frequency di-vision multiplexing system.

Arthur James Lowery (M’91–SM’96–F’09) wasborn in Yorkshire, England, in 1961. He receivedthe B.Sc. degree (first class hons) in applied physicsfrom the University of Durham, U.K., in 1983and the Ph.D. degree in electrical and electronicengineering from the University of Nottingham,Nottingham, U.K., in 1988.

In 1984 he joined the University of Nottinghamas a lecturer and pioneered time-domain field mod-elling of semiconductor lasers as the Transmission-Line Laser Model. In 1990 he emigrated to Australia

to become a Senior Lecturer at the Photonics Research Laboratory at the Uni-versity of Melbourne. After working on Photonic-CAD, Packet Switching andLaser Ranging, he was promoted to Associate Professor and Reader in 1993.He continued to develop novel time-domain simulation techniques and in 1996he co-founded Virtual Photonics Pty Ltd. Virtual Photonics merged with BNeD(Berlin) in 1998, with Lowery as CTO, leading the development of VPI’s phys-ical-level photonic design automation tools such as VPItransmissionMaker andVPIcomponentMaker, which are widely used in industry and academia for de-velopment, research and teaching. In 2004 he was appointed Professor of Elec-trical and Computer Systems Engineering at Monash University, Melbourne,Australia where he is working on the applications of photonic technology, in-cluding a bionic vision device. In 2008 he founded Ofidium Pty Ltd to com-mercialise optical OFDM. He has published more than 200 papers and 4 bookchapters on the simulation of photonic devices and circuits and photonic appli-cations such as mode-locking, packet switching, OFDM transmission systemsand high-speed photonic circuits.

Prof. Lowery is a Fellow of the IEEE, the IET (UK) and the AustralianAcademy of Technological Sciences and Engineering (ATSE), from which hereceived a 2007 Clunies-Ross Award for his work on photonic simulation.